In order to investigate how a streaming potential coefficient measured in the laboratory, at a typical scale of 10 cm, can be incorporated into a field model, with a typical scale of 1 to 10 km, we measured the electric field induced by water flows forced at 150 m depth through a 10-m wide granite fractured zone. The water flows were obtained by pumping cyclically 10 m of water from a borehole that cut the fractured zone at depth, and contemporaneously reinjecting it into another borehole located 50 m away. After one day a steady-state fluid flow regime was reached, with pumping cycles lasting 45 minutes, indicating a hydraulic conductivity of 10 À5 m s À1 and a specific storage coefficient of 3:25 Â 10 À6 m À1 . The expected self-potential at the surface was an anomaly with two maxima of opposite sign and 2 lV amplitude each, both located 160 m away from the middle of the borehole heads, the signal being divided by two 500 m away from the middle of the borehole heads (in agreement with WURMSTICH and MORGAN, 1994). Instead, we observed an electrical signal of 8 mV midway between the borehole heads, and smaller than 5 mV, 33 m away from the borehole heads. The discrepancy observed between the data and the model can be explained by fluid flow leakages that occurred close to the water-table head, represented about 20% of the total water flow, and activated smaller but closer electric sources. This experiment thus illustrates the practical difficulty of detecting streaming potentials generated at depth. It shows in particular that in fractured zones, and hence in the vicinity of a major active fault small water flows located distantly from an energetic targeted source, but close to some of the electrodes of the network, can sometimes drastically distort the shape of the expected anomaly. Models of possible electrical earthquake precursors therefore turn out to be more speculative than expected.